Characterizing the Observation Bias in Gravitational-wave Detections and Finding Structured Population Properties

Doğa Veske, I. Bartos, Zsuzsa Márka, Szabolcs MárkaThe Astrophysical Journal 2021 · DOI

TL;DR

This paper tackles the "Malmquist bias" in gravitational-wave (GW) detections, where easier-to-detect events skew observed source populations. They introduce a computationally efficient semi-analytical method to estimate this bias, achieving similar accuracy to traditional simulations but at a fraction of the cost. This method is then used to uncover previously unseen structures in the binary black hole (BBH) mass distribution.

What Problem Does This Paper Address?

Gravitational-wave detectors like LIGO and Virgo are not equally sensitive to all astrophysical events. Events that are intrinsically brighter or closer to Earth are more likely to be detected. This leads to an "observation bias," where the observed distribution of source properties (like black hole masses) doesn't accurately reflect the true underlying astrophysical distribution. This is analogous to the Malmquist bias in astronomy. Accurately characterizing this bias is crucial for understanding the formation channels and evolution of compact objects, but current methods relying on extensive numerical simulations are computationally expensive.

Methods

The authors develop an efficient semi-analytical method to estimate the observation bias. Instead of simulating millions of BBH mergers, they derive analytical expressions for the signal-to-noise ratio (SNR) as a function of source properties and detector characteristics. They then use a numerically computed list of SNRs to evaluate these expressions. This approach is significantly faster than traditional injection-based simulations. The method is validated against simulations and then applied to infer BBH population properties.

Datasets

The study utilizes data from the Gravitational Wave Transient Catalogues GWTC-1 and GWTC-2, which contain detections of BBH mergers. For validation, they reference simulation studies that inject BBH merger waveforms (specifically mentioning IMRPhenomPv2 and SEOBNRv4) into simulated noise.

Key Insights and Ideas

  • Bias is fundamental: The paper emphasizes that observation bias is an inherent consequence of how GW detectors work and is analogous to well-known biases in other astronomical observations.
  • Computational efficiency is key: The development of a semi-analytical method offers a significant speedup (estimated at 10^6 times faster than traditional simulations), making bias correction more accessible.
  • Uncovering hidden structures: By correcting for observation bias, the authors reveal previously undetected structures in the BBH mass distribution, suggesting that current models might be too simplistic.
  • Two distinct BBH populations: The analysis points to two prominent peaks in the BBH mass distribution, around 10 M☉ and 30 M☉, which could indicate different formation pathways.

Key Results

The developed semi-analytical method for estimating observation bias is shown to be as accurate as traditional numerical simulations but computationally orders of magnitude faster. Applying this method to GWTC-1 and GWTC-2 data reveals two significant peaks in the BBH mass distribution, with mean posterior densities around these peaks being 3.3 and 2.0 times higher than the mean prior densities. The study also notes a sharper decrease in the posterior mean at higher masses, potentially indicating the presence of a pair-instability mass gap.

Limitations and Future Work

The current study focuses on non-spinning, quasicircular BBH mergers. The authors acknowledge that the effect of spin on the observation bias is not considered and suggest this as a direction for future work. Additionally, the analysis of the BBH population is based on a histogram-type approach, and further work could explore more sophisticated models. The authors also share their code and datasets, encouraging further research and application of their method.

Summary updated

Abstract

Abstract The observed distributions of the source properties from gravitational-wave (GW) detections are biased due to the selection effects and detection criteria in the detections, analogous to the Malmquist bias. In this work, this observation bias is investigated through its fundamental statistical and physical origins. An efficient semi-analytical formulation for its estimation is derived, which is as accurate as the standard method of numerical simulations, with only a millionth of the computational cost. Then, the estimated bias is used for unmodeled inferences on the binary black hole population. These inferences show additional structures, specifically two peaks in the joint mass distribution around binary masses ∼10 M ⊙ and ∼30 M ⊙ . Example ready-to-use scripts and some produced data sets for this method are shared in an online repository.
BibTeX
@article{veske2021characterizing,
  author = {Doğa Veske and I. Bartos and Zsuzsa Márka and Szabolcs Márka},
  title = {{Characterizing the Observation Bias in Gravitational-wave Detections and Finding Structured Population Properties}},
  journal = {The Astrophysical Journal},
  year = {2021},
  doi = {10.3847/1538-4357/ac27ac},
}
Figure